{"title":"Slice-based YOLO-SAM hybrid prompting framework for automatic medical image segmentation.","authors":"Xiaobo Liu, Qiuyan Pei, Xiaoyu Wang, Honghang Luo","doi":"10.1371/journal.pone.0360939","DOIUrl":null,"url":null,"abstract":"<p><p>Accurate medical image segmentation is essential for computer-aided diagnosis, treatment planning, and quantitative image analysis. This study evaluates a slice-based automatic segmentation framework that couples YOLOv8-based region localization with prompt-driven Segment Anything Model (SAM) segmentation. In contrast to a simple YOLOv8 + SAM cascade using only bounding-box prompts, the proposed framework adds hybrid prompt generation with bounding boxes and positive/negative points, parameter-efficient fine-tuning of the SAM prompt encoder and mask decoder, and lightweight post-processing. The framework is evaluated on two public datasets under explicitly defined local protocols: 2D slice-based prostate MRI segmentation using Prostate158 and skin lesion segmentation using an internal ISIC 2018 split. On Prostate158, the method obtains Dice coefficients of 89.2%, 76.8%, and 65.3% for central gland, peripheral zone, and prostate cancer lesion segmentation, respectively. On the local ISIC 2018 split, it obtains a Dice coefficient of 94.8% and a Jaccard index of 91.5%. These results indicate competitive performance under the implemented experimental protocol, but they should not be interpreted as protocol-independent state-of-the-art claims. The manuscript clarifies the evaluation unit, prompt-source protocol, 2D processing strategy, hyperparameter settings, baseline grouping, and limitations related to statistical uncertainty, detector failure propagation, and clinical translation.</p>","PeriodicalId":20189,"journal":{"name":"PLoS ONE","volume":"21 11","pages":"e0360939"},"PeriodicalIF":2.8000,"publicationDate":"2026-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13766228/pdf/","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"PLoS ONE","FirstCategoryId":"103","ListUrlMain":"https://doi.org/10.1371/journal.pone.0360939","RegionNum":3,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"2026/1/1 0:00:00","PubModel":"eCollection","JCR":"Q1","JCRName":"MULTIDISCIPLINARY SCIENCES","Score":null,"Total":0}
引用次数: 0
Abstract
Accurate medical image segmentation is essential for computer-aided diagnosis, treatment planning, and quantitative image analysis. This study evaluates a slice-based automatic segmentation framework that couples YOLOv8-based region localization with prompt-driven Segment Anything Model (SAM) segmentation. In contrast to a simple YOLOv8 + SAM cascade using only bounding-box prompts, the proposed framework adds hybrid prompt generation with bounding boxes and positive/negative points, parameter-efficient fine-tuning of the SAM prompt encoder and mask decoder, and lightweight post-processing. The framework is evaluated on two public datasets under explicitly defined local protocols: 2D slice-based prostate MRI segmentation using Prostate158 and skin lesion segmentation using an internal ISIC 2018 split. On Prostate158, the method obtains Dice coefficients of 89.2%, 76.8%, and 65.3% for central gland, peripheral zone, and prostate cancer lesion segmentation, respectively. On the local ISIC 2018 split, it obtains a Dice coefficient of 94.8% and a Jaccard index of 91.5%. These results indicate competitive performance under the implemented experimental protocol, but they should not be interpreted as protocol-independent state-of-the-art claims. The manuscript clarifies the evaluation unit, prompt-source protocol, 2D processing strategy, hyperparameter settings, baseline grouping, and limitations related to statistical uncertainty, detector failure propagation, and clinical translation.
期刊介绍:
PLOS ONE is an international, peer-reviewed, open-access, online publication. PLOS ONE welcomes reports on primary research from any scientific discipline. It provides:
* Open-access—freely accessible online, authors retain copyright
* Fast publication times
* Peer review by expert, practicing researchers
* Post-publication tools to indicate quality and impact
* Community-based dialogue on articles
* Worldwide media coverage